Commit | Line | Data |
---|---|---|
470decc6 | 1 | /* |
58862699 | 2 | * linux/fs/jbd2/checkpoint.c |
470decc6 DK |
3 | * |
4 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1999 | |
5 | * | |
6 | * Copyright 1999 Red Hat Software --- All Rights Reserved | |
7 | * | |
8 | * This file is part of the Linux kernel and is made available under | |
9 | * the terms of the GNU General Public License, version 2, or at your | |
10 | * option, any later version, incorporated herein by reference. | |
11 | * | |
12 | * Checkpoint routines for the generic filesystem journaling code. | |
13 | * Part of the ext2fs journaling system. | |
14 | * | |
15 | * Checkpointing is the process of ensuring that a section of the log is | |
16 | * committed fully to disk, so that that portion of the log can be | |
17 | * reused. | |
18 | */ | |
19 | ||
20 | #include <linux/time.h> | |
21 | #include <linux/fs.h> | |
f7f4bccb | 22 | #include <linux/jbd2.h> |
470decc6 DK |
23 | #include <linux/errno.h> |
24 | #include <linux/slab.h> | |
cc3e1bea | 25 | #include <linux/blkdev.h> |
879c5e6b | 26 | #include <trace/events/jbd2.h> |
470decc6 DK |
27 | |
28 | /* | |
29 | * Unlink a buffer from a transaction checkpoint list. | |
30 | * | |
31 | * Called with j_list_lock held. | |
32 | */ | |
33 | static inline void __buffer_unlink_first(struct journal_head *jh) | |
34 | { | |
35 | transaction_t *transaction = jh->b_cp_transaction; | |
36 | ||
37 | jh->b_cpnext->b_cpprev = jh->b_cpprev; | |
38 | jh->b_cpprev->b_cpnext = jh->b_cpnext; | |
39 | if (transaction->t_checkpoint_list == jh) { | |
40 | transaction->t_checkpoint_list = jh->b_cpnext; | |
41 | if (transaction->t_checkpoint_list == jh) | |
42 | transaction->t_checkpoint_list = NULL; | |
43 | } | |
44 | } | |
45 | ||
46 | /* | |
47 | * Unlink a buffer from a transaction checkpoint(io) list. | |
48 | * | |
49 | * Called with j_list_lock held. | |
50 | */ | |
51 | static inline void __buffer_unlink(struct journal_head *jh) | |
52 | { | |
53 | transaction_t *transaction = jh->b_cp_transaction; | |
54 | ||
55 | __buffer_unlink_first(jh); | |
56 | if (transaction->t_checkpoint_io_list == jh) { | |
57 | transaction->t_checkpoint_io_list = jh->b_cpnext; | |
58 | if (transaction->t_checkpoint_io_list == jh) | |
59 | transaction->t_checkpoint_io_list = NULL; | |
60 | } | |
61 | } | |
62 | ||
63 | /* | |
64 | * Move a buffer from the checkpoint list to the checkpoint io list | |
65 | * | |
66 | * Called with j_list_lock held | |
67 | */ | |
68 | static inline void __buffer_relink_io(struct journal_head *jh) | |
69 | { | |
70 | transaction_t *transaction = jh->b_cp_transaction; | |
71 | ||
72 | __buffer_unlink_first(jh); | |
73 | ||
74 | if (!transaction->t_checkpoint_io_list) { | |
75 | jh->b_cpnext = jh->b_cpprev = jh; | |
76 | } else { | |
77 | jh->b_cpnext = transaction->t_checkpoint_io_list; | |
78 | jh->b_cpprev = transaction->t_checkpoint_io_list->b_cpprev; | |
79 | jh->b_cpprev->b_cpnext = jh; | |
80 | jh->b_cpnext->b_cpprev = jh; | |
81 | } | |
82 | transaction->t_checkpoint_io_list = jh; | |
83 | } | |
84 | ||
85 | /* | |
86 | * Try to release a checkpointed buffer from its transaction. | |
87 | * Returns 1 if we released it and 2 if we also released the | |
88 | * whole transaction. | |
89 | * | |
90 | * Requires j_list_lock | |
470decc6 DK |
91 | */ |
92 | static int __try_to_free_cp_buf(struct journal_head *jh) | |
93 | { | |
94 | int ret = 0; | |
95 | struct buffer_head *bh = jh2bh(jh); | |
96 | ||
932bb305 | 97 | if (jh->b_transaction == NULL && !buffer_locked(bh) && |
44519faf | 98 | !buffer_dirty(bh) && !buffer_write_io_error(bh)) { |
de1b7941 JK |
99 | /* |
100 | * Get our reference so that bh cannot be freed before | |
101 | * we unlock it | |
102 | */ | |
103 | get_bh(bh); | |
470decc6 | 104 | JBUFFER_TRACE(jh, "remove from checkpoint list"); |
f7f4bccb | 105 | ret = __jbd2_journal_remove_checkpoint(jh) + 1; |
470decc6 DK |
106 | BUFFER_TRACE(bh, "release"); |
107 | __brelse(bh); | |
470decc6 DK |
108 | } |
109 | return ret; | |
110 | } | |
111 | ||
112 | /* | |
f7f4bccb | 113 | * __jbd2_log_wait_for_space: wait until there is space in the journal. |
470decc6 DK |
114 | * |
115 | * Called under j-state_lock *only*. It will be unlocked if we have to wait | |
116 | * for a checkpoint to free up some space in the log. | |
117 | */ | |
f7f4bccb | 118 | void __jbd2_log_wait_for_space(journal_t *journal) |
470decc6 | 119 | { |
8c3f25d8 | 120 | int nblocks, space_left; |
a931da6a | 121 | /* assert_spin_locked(&journal->j_state_lock); */ |
470decc6 | 122 | |
76c39904 JK |
123 | nblocks = jbd2_space_needed(journal); |
124 | while (jbd2_log_space_left(journal) < nblocks) { | |
f7f4bccb | 125 | if (journal->j_flags & JBD2_ABORT) |
470decc6 | 126 | return; |
a931da6a | 127 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
128 | mutex_lock(&journal->j_checkpoint_mutex); |
129 | ||
130 | /* | |
131 | * Test again, another process may have checkpointed while we | |
23f8b79e | 132 | * were waiting for the checkpoint lock. If there are no |
8c3f25d8 TT |
133 | * transactions ready to be checkpointed, try to recover |
134 | * journal space by calling cleanup_journal_tail(), and if | |
135 | * that doesn't work, by waiting for the currently committing | |
136 | * transaction to complete. If there is absolutely no way | |
137 | * to make progress, this is either a BUG or corrupted | |
138 | * filesystem, so abort the journal and leave a stack | |
139 | * trace for forensic evidence. | |
470decc6 | 140 | */ |
a931da6a | 141 | write_lock(&journal->j_state_lock); |
23f8b79e | 142 | spin_lock(&journal->j_list_lock); |
76c39904 JK |
143 | nblocks = jbd2_space_needed(journal); |
144 | space_left = jbd2_log_space_left(journal); | |
8c3f25d8 | 145 | if (space_left < nblocks) { |
23f8b79e | 146 | int chkpt = journal->j_checkpoint_transactions != NULL; |
8c3f25d8 | 147 | tid_t tid = 0; |
23f8b79e | 148 | |
8c3f25d8 TT |
149 | if (journal->j_committing_transaction) |
150 | tid = journal->j_committing_transaction->t_tid; | |
23f8b79e | 151 | spin_unlock(&journal->j_list_lock); |
a931da6a | 152 | write_unlock(&journal->j_state_lock); |
23f8b79e DG |
153 | if (chkpt) { |
154 | jbd2_log_do_checkpoint(journal); | |
8c3f25d8 TT |
155 | } else if (jbd2_cleanup_journal_tail(journal) == 0) { |
156 | /* We were able to recover space; yay! */ | |
157 | ; | |
158 | } else if (tid) { | |
0ef54180 PG |
159 | /* |
160 | * jbd2_journal_commit_transaction() may want | |
161 | * to take the checkpoint_mutex if JBD2_FLUSHED | |
162 | * is set. So we need to temporarily drop it. | |
163 | */ | |
164 | mutex_unlock(&journal->j_checkpoint_mutex); | |
8c3f25d8 | 165 | jbd2_log_wait_commit(journal, tid); |
0ef54180 PG |
166 | write_lock(&journal->j_state_lock); |
167 | continue; | |
23f8b79e | 168 | } else { |
8c3f25d8 TT |
169 | printk(KERN_ERR "%s: needed %d blocks and " |
170 | "only had %d space available\n", | |
171 | __func__, nblocks, space_left); | |
172 | printk(KERN_ERR "%s: no way to get more " | |
173 | "journal space in %s\n", __func__, | |
174 | journal->j_devname); | |
175 | WARN_ON(1); | |
23f8b79e DG |
176 | jbd2_journal_abort(journal, 0); |
177 | } | |
a931da6a | 178 | write_lock(&journal->j_state_lock); |
23f8b79e DG |
179 | } else { |
180 | spin_unlock(&journal->j_list_lock); | |
470decc6 DK |
181 | } |
182 | mutex_unlock(&journal->j_checkpoint_mutex); | |
183 | } | |
184 | } | |
185 | ||
470decc6 DK |
186 | /* |
187 | * Clean up transaction's list of buffers submitted for io. | |
188 | * We wait for any pending IO to complete and remove any clean | |
189 | * buffers. Note that we take the buffers in the opposite ordering | |
190 | * from the one in which they were submitted for IO. | |
191 | * | |
44519faf HK |
192 | * Return 0 on success, and return <0 if some buffers have failed |
193 | * to be written out. | |
194 | * | |
470decc6 DK |
195 | * Called with j_list_lock held. |
196 | */ | |
44519faf | 197 | static int __wait_cp_io(journal_t *journal, transaction_t *transaction) |
470decc6 DK |
198 | { |
199 | struct journal_head *jh; | |
200 | struct buffer_head *bh; | |
201 | tid_t this_tid; | |
202 | int released = 0; | |
44519faf | 203 | int ret = 0; |
470decc6 DK |
204 | |
205 | this_tid = transaction->t_tid; | |
206 | restart: | |
207 | /* Did somebody clean up the transaction in the meanwhile? */ | |
208 | if (journal->j_checkpoint_transactions != transaction || | |
209 | transaction->t_tid != this_tid) | |
44519faf | 210 | return ret; |
470decc6 DK |
211 | while (!released && transaction->t_checkpoint_io_list) { |
212 | jh = transaction->t_checkpoint_io_list; | |
213 | bh = jh2bh(jh); | |
de1b7941 | 214 | get_bh(bh); |
470decc6 | 215 | if (buffer_locked(bh)) { |
470decc6 | 216 | spin_unlock(&journal->j_list_lock); |
470decc6 DK |
217 | wait_on_buffer(bh); |
218 | /* the journal_head may have gone by now */ | |
219 | BUFFER_TRACE(bh, "brelse"); | |
220 | __brelse(bh); | |
221 | spin_lock(&journal->j_list_lock); | |
222 | goto restart; | |
223 | } | |
44519faf HK |
224 | if (unlikely(buffer_write_io_error(bh))) |
225 | ret = -EIO; | |
226 | ||
470decc6 DK |
227 | /* |
228 | * Now in whatever state the buffer currently is, we know that | |
229 | * it has been written out and so we can drop it from the list | |
230 | */ | |
f7f4bccb | 231 | released = __jbd2_journal_remove_checkpoint(jh); |
470decc6 DK |
232 | __brelse(bh); |
233 | } | |
44519faf HK |
234 | |
235 | return ret; | |
470decc6 DK |
236 | } |
237 | ||
470decc6 | 238 | static void |
1a0d3786 | 239 | __flush_batch(journal_t *journal, int *batch_count) |
470decc6 DK |
240 | { |
241 | int i; | |
d3ad8434 | 242 | struct blk_plug plug; |
470decc6 | 243 | |
d3ad8434 | 244 | blk_start_plug(&plug); |
9cb569d6 | 245 | for (i = 0; i < *batch_count; i++) |
d3ad8434 TM |
246 | write_dirty_buffer(journal->j_chkpt_bhs[i], WRITE_SYNC); |
247 | blk_finish_plug(&plug); | |
9cb569d6 | 248 | |
470decc6 | 249 | for (i = 0; i < *batch_count; i++) { |
1a0d3786 | 250 | struct buffer_head *bh = journal->j_chkpt_bhs[i]; |
470decc6 DK |
251 | BUFFER_TRACE(bh, "brelse"); |
252 | __brelse(bh); | |
253 | } | |
254 | *batch_count = 0; | |
255 | } | |
256 | ||
470decc6 DK |
257 | /* |
258 | * Perform an actual checkpoint. We take the first transaction on the | |
259 | * list of transactions to be checkpointed and send all its buffers | |
260 | * to disk. We submit larger chunks of data at once. | |
261 | * | |
262 | * The journal should be locked before calling this function. | |
44519faf | 263 | * Called with j_checkpoint_mutex held. |
470decc6 | 264 | */ |
f7f4bccb | 265 | int jbd2_log_do_checkpoint(journal_t *journal) |
470decc6 | 266 | { |
be1158cc TT |
267 | struct journal_head *jh; |
268 | struct buffer_head *bh; | |
269 | transaction_t *transaction; | |
270 | tid_t this_tid; | |
271 | int err, result, batch_count = 0; | |
470decc6 DK |
272 | |
273 | jbd_debug(1, "Start checkpoint\n"); | |
274 | ||
275 | /* | |
276 | * First thing: if there are any transactions in the log which | |
277 | * don't need checkpointing, just eliminate them from the | |
278 | * journal straight away. | |
279 | */ | |
f7f4bccb | 280 | result = jbd2_cleanup_journal_tail(journal); |
879c5e6b | 281 | trace_jbd2_checkpoint(journal, result); |
470decc6 DK |
282 | jbd_debug(1, "cleanup_journal_tail returned %d\n", result); |
283 | if (result <= 0) | |
284 | return result; | |
285 | ||
286 | /* | |
287 | * OK, we need to start writing disk blocks. Take one transaction | |
288 | * and write it. | |
289 | */ | |
44519faf | 290 | result = 0; |
470decc6 DK |
291 | spin_lock(&journal->j_list_lock); |
292 | if (!journal->j_checkpoint_transactions) | |
293 | goto out; | |
294 | transaction = journal->j_checkpoint_transactions; | |
8e85fb3f JL |
295 | if (transaction->t_chp_stats.cs_chp_time == 0) |
296 | transaction->t_chp_stats.cs_chp_time = jiffies; | |
470decc6 DK |
297 | this_tid = transaction->t_tid; |
298 | restart: | |
299 | /* | |
300 | * If someone cleaned up this transaction while we slept, we're | |
301 | * done (maybe it's a new transaction, but it fell at the same | |
302 | * address). | |
303 | */ | |
be1158cc TT |
304 | if (journal->j_checkpoint_transactions != transaction || |
305 | transaction->t_tid != this_tid) | |
306 | goto out; | |
470decc6 | 307 | |
be1158cc TT |
308 | /* checkpoint all of the transaction's buffers */ |
309 | while (transaction->t_checkpoint_list) { | |
310 | jh = transaction->t_checkpoint_list; | |
311 | bh = jh2bh(jh); | |
312 | ||
313 | if (buffer_locked(bh)) { | |
314 | spin_unlock(&journal->j_list_lock); | |
315 | get_bh(bh); | |
316 | wait_on_buffer(bh); | |
317 | /* the journal_head may have gone by now */ | |
318 | BUFFER_TRACE(bh, "brelse"); | |
319 | __brelse(bh); | |
320 | goto retry; | |
470decc6 | 321 | } |
be1158cc TT |
322 | if (jh->b_transaction != NULL) { |
323 | transaction_t *t = jh->b_transaction; | |
324 | tid_t tid = t->t_tid; | |
470decc6 | 325 | |
be1158cc TT |
326 | transaction->t_chp_stats.cs_forced_to_close++; |
327 | spin_unlock(&journal->j_list_lock); | |
328 | if (unlikely(journal->j_flags & JBD2_UNMOUNT)) | |
329 | /* | |
330 | * The journal thread is dead; so | |
331 | * starting and waiting for a commit | |
332 | * to finish will cause us to wait for | |
333 | * a _very_ long time. | |
334 | */ | |
335 | printk(KERN_ERR | |
336 | "JBD2: %s: Waiting for Godot: block %llu\n", | |
337 | journal->j_devname, (unsigned long long) bh->b_blocknr); | |
338 | ||
339 | jbd2_log_start_commit(journal, tid); | |
340 | jbd2_log_wait_commit(journal, tid); | |
341 | goto retry; | |
342 | } | |
343 | if (!buffer_dirty(bh)) { | |
344 | if (unlikely(buffer_write_io_error(bh)) && !result) | |
345 | result = -EIO; | |
346 | get_bh(bh); | |
347 | BUFFER_TRACE(bh, "remove from checkpoint"); | |
348 | __jbd2_journal_remove_checkpoint(jh); | |
349 | spin_unlock(&journal->j_list_lock); | |
350 | __brelse(bh); | |
351 | goto retry; | |
470decc6 DK |
352 | } |
353 | /* | |
be1158cc TT |
354 | * Important: we are about to write the buffer, and |
355 | * possibly block, while still holding the journal | |
356 | * lock. We cannot afford to let the transaction | |
357 | * logic start messing around with this buffer before | |
358 | * we write it to disk, as that would break | |
359 | * recoverability. | |
470decc6 | 360 | */ |
be1158cc TT |
361 | BUFFER_TRACE(bh, "queue"); |
362 | get_bh(bh); | |
363 | J_ASSERT_BH(bh, !buffer_jwrite(bh)); | |
364 | journal->j_chkpt_bhs[batch_count++] = bh; | |
365 | __buffer_relink_io(jh); | |
366 | transaction->t_chp_stats.cs_written++; | |
367 | if ((batch_count == JBD2_NR_BATCH) || | |
368 | need_resched() || | |
369 | spin_needbreak(&journal->j_list_lock)) | |
370 | goto unlock_and_flush; | |
470decc6 | 371 | } |
be1158cc TT |
372 | |
373 | if (batch_count) { | |
374 | unlock_and_flush: | |
375 | spin_unlock(&journal->j_list_lock); | |
376 | retry: | |
377 | if (batch_count) | |
378 | __flush_batch(journal, &batch_count); | |
379 | spin_lock(&journal->j_list_lock); | |
380 | goto restart; | |
381 | } | |
382 | ||
383 | /* | |
384 | * Now we issued all of the transaction's buffers, let's deal | |
385 | * with the buffers that are out for I/O. | |
386 | */ | |
387 | err = __wait_cp_io(journal, transaction); | |
388 | if (!result) | |
389 | result = err; | |
470decc6 DK |
390 | out: |
391 | spin_unlock(&journal->j_list_lock); | |
470decc6 | 392 | if (result < 0) |
44519faf HK |
393 | jbd2_journal_abort(journal, result); |
394 | else | |
395 | result = jbd2_cleanup_journal_tail(journal); | |
396 | ||
397 | return (result < 0) ? result : 0; | |
470decc6 DK |
398 | } |
399 | ||
400 | /* | |
401 | * Check the list of checkpoint transactions for the journal to see if | |
402 | * we have already got rid of any since the last update of the log tail | |
403 | * in the journal superblock. If so, we can instantly roll the | |
404 | * superblock forward to remove those transactions from the log. | |
405 | * | |
406 | * Return <0 on error, 0 on success, 1 if there was nothing to clean up. | |
407 | * | |
408 | * Called with the journal lock held. | |
409 | * | |
410 | * This is the only part of the journaling code which really needs to be | |
411 | * aware of transaction aborts. Checkpointing involves writing to the | |
412 | * main filesystem area rather than to the journal, so it can proceed | |
44519faf HK |
413 | * even in abort state, but we must not update the super block if |
414 | * checkpointing may have failed. Otherwise, we would lose some metadata | |
415 | * buffers which should be written-back to the filesystem. | |
470decc6 DK |
416 | */ |
417 | ||
f7f4bccb | 418 | int jbd2_cleanup_journal_tail(journal_t *journal) |
470decc6 | 419 | { |
470decc6 | 420 | tid_t first_tid; |
79feb521 | 421 | unsigned long blocknr; |
470decc6 | 422 | |
44519faf HK |
423 | if (is_journal_aborted(journal)) |
424 | return 1; | |
425 | ||
79feb521 | 426 | if (!jbd2_journal_get_log_tail(journal, &first_tid, &blocknr)) |
470decc6 | 427 | return 1; |
79feb521 | 428 | J_ASSERT(blocknr != 0); |
cc3e1bea TT |
429 | |
430 | /* | |
79feb521 JK |
431 | * We need to make sure that any blocks that were recently written out |
432 | * --- perhaps by jbd2_log_do_checkpoint() --- are flushed out before | |
433 | * we drop the transactions from the journal. It's unlikely this will | |
434 | * be necessary, especially with an appropriately sized journal, but we | |
435 | * need this to guarantee correctness. Fortunately | |
436 | * jbd2_cleanup_journal_tail() doesn't get called all that often. | |
cc3e1bea | 437 | */ |
79feb521 | 438 | if (journal->j_flags & JBD2_BARRIER) |
dd3932ed | 439 | blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL); |
79feb521 JK |
440 | |
441 | __jbd2_update_log_tail(journal, first_tid, blocknr); | |
470decc6 DK |
442 | return 0; |
443 | } | |
444 | ||
445 | ||
446 | /* Checkpoint list management */ | |
447 | ||
448 | /* | |
449 | * journal_clean_one_cp_list | |
450 | * | |
de1b7941 JK |
451 | * Find all the written-back checkpoint buffers in the given list and |
452 | * release them. | |
470decc6 DK |
453 | * |
454 | * Called with the journal locked. | |
455 | * Called with j_list_lock held. | |
90802ed9 | 456 | * Returns number of buffers reaped (for debug) |
470decc6 DK |
457 | */ |
458 | ||
459 | static int journal_clean_one_cp_list(struct journal_head *jh, int *released) | |
460 | { | |
461 | struct journal_head *last_jh; | |
462 | struct journal_head *next_jh = jh; | |
463 | int ret, freed = 0; | |
464 | ||
465 | *released = 0; | |
466 | if (!jh) | |
467 | return 0; | |
468 | ||
469 | last_jh = jh->b_cpprev; | |
470 | do { | |
471 | jh = next_jh; | |
472 | next_jh = jh->b_cpnext; | |
932bb305 JK |
473 | ret = __try_to_free_cp_buf(jh); |
474 | if (ret) { | |
475 | freed++; | |
476 | if (ret == 2) { | |
477 | *released = 1; | |
478 | return freed; | |
470decc6 DK |
479 | } |
480 | } | |
481 | /* | |
482 | * This function only frees up some memory | |
483 | * if possible so we dont have an obligation | |
484 | * to finish processing. Bail out if preemption | |
485 | * requested: | |
486 | */ | |
487 | if (need_resched()) | |
488 | return freed; | |
489 | } while (jh != last_jh); | |
490 | ||
491 | return freed; | |
492 | } | |
493 | ||
494 | /* | |
495 | * journal_clean_checkpoint_list | |
496 | * | |
497 | * Find all the written-back checkpoint buffers in the journal and release them. | |
498 | * | |
499 | * Called with the journal locked. | |
500 | * Called with j_list_lock held. | |
501 | * Returns number of buffers reaped (for debug) | |
502 | */ | |
503 | ||
f7f4bccb | 504 | int __jbd2_journal_clean_checkpoint_list(journal_t *journal) |
470decc6 DK |
505 | { |
506 | transaction_t *transaction, *last_transaction, *next_transaction; | |
507 | int ret = 0; | |
508 | int released; | |
509 | ||
510 | transaction = journal->j_checkpoint_transactions; | |
511 | if (!transaction) | |
512 | goto out; | |
513 | ||
514 | last_transaction = transaction->t_cpprev; | |
515 | next_transaction = transaction; | |
516 | do { | |
517 | transaction = next_transaction; | |
518 | next_transaction = transaction->t_cpnext; | |
519 | ret += journal_clean_one_cp_list(transaction-> | |
520 | t_checkpoint_list, &released); | |
521 | /* | |
522 | * This function only frees up some memory if possible so we | |
523 | * dont have an obligation to finish processing. Bail out if | |
524 | * preemption requested: | |
525 | */ | |
526 | if (need_resched()) | |
527 | goto out; | |
528 | if (released) | |
529 | continue; | |
530 | /* | |
531 | * It is essential that we are as careful as in the case of | |
532 | * t_checkpoint_list with removing the buffer from the list as | |
533 | * we can possibly see not yet submitted buffers on io_list | |
534 | */ | |
535 | ret += journal_clean_one_cp_list(transaction-> | |
536 | t_checkpoint_io_list, &released); | |
537 | if (need_resched()) | |
538 | goto out; | |
539 | } while (transaction != last_transaction); | |
540 | out: | |
541 | return ret; | |
542 | } | |
543 | ||
544 | /* | |
545 | * journal_remove_checkpoint: called after a buffer has been committed | |
546 | * to disk (either by being write-back flushed to disk, or being | |
547 | * committed to the log). | |
548 | * | |
549 | * We cannot safely clean a transaction out of the log until all of the | |
550 | * buffer updates committed in that transaction have safely been stored | |
551 | * elsewhere on disk. To achieve this, all of the buffers in a | |
552 | * transaction need to be maintained on the transaction's checkpoint | |
553 | * lists until they have been rewritten, at which point this function is | |
554 | * called to remove the buffer from the existing transaction's | |
555 | * checkpoint lists. | |
556 | * | |
557 | * The function returns 1 if it frees the transaction, 0 otherwise. | |
de1b7941 | 558 | * The function can free jh and bh. |
470decc6 | 559 | * |
470decc6 | 560 | * This function is called with j_list_lock held. |
470decc6 | 561 | */ |
f7f4bccb | 562 | int __jbd2_journal_remove_checkpoint(struct journal_head *jh) |
470decc6 | 563 | { |
bf699327 | 564 | struct transaction_chp_stats_s *stats; |
470decc6 DK |
565 | transaction_t *transaction; |
566 | journal_t *journal; | |
567 | int ret = 0; | |
568 | ||
569 | JBUFFER_TRACE(jh, "entry"); | |
570 | ||
571 | if ((transaction = jh->b_cp_transaction) == NULL) { | |
572 | JBUFFER_TRACE(jh, "not on transaction"); | |
573 | goto out; | |
574 | } | |
575 | journal = transaction->t_journal; | |
576 | ||
de1b7941 | 577 | JBUFFER_TRACE(jh, "removing from transaction"); |
470decc6 DK |
578 | __buffer_unlink(jh); |
579 | jh->b_cp_transaction = NULL; | |
de1b7941 | 580 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
581 | |
582 | if (transaction->t_checkpoint_list != NULL || | |
583 | transaction->t_checkpoint_io_list != NULL) | |
584 | goto out; | |
470decc6 DK |
585 | |
586 | /* | |
587 | * There is one special case to worry about: if we have just pulled the | |
f5a7a6b0 JK |
588 | * buffer off a running or committing transaction's checkpoing list, |
589 | * then even if the checkpoint list is empty, the transaction obviously | |
590 | * cannot be dropped! | |
470decc6 | 591 | * |
f5a7a6b0 | 592 | * The locking here around t_state is a bit sleazy. |
f7f4bccb | 593 | * See the comment at the end of jbd2_journal_commit_transaction(). |
470decc6 | 594 | */ |
de1b7941 | 595 | if (transaction->t_state != T_FINISHED) |
470decc6 | 596 | goto out; |
470decc6 DK |
597 | |
598 | /* OK, that was the last buffer for the transaction: we can now | |
599 | safely remove this transaction from the log */ | |
bf699327 TT |
600 | stats = &transaction->t_chp_stats; |
601 | if (stats->cs_chp_time) | |
602 | stats->cs_chp_time = jbd2_time_diff(stats->cs_chp_time, | |
603 | jiffies); | |
604 | trace_jbd2_checkpoint_stats(journal->j_fs_dev->bd_dev, | |
605 | transaction->t_tid, stats); | |
470decc6 | 606 | |
f7f4bccb | 607 | __jbd2_journal_drop_transaction(journal, transaction); |
0c2022ec | 608 | jbd2_journal_free_transaction(transaction); |
470decc6 DK |
609 | ret = 1; |
610 | out: | |
470decc6 DK |
611 | return ret; |
612 | } | |
613 | ||
614 | /* | |
615 | * journal_insert_checkpoint: put a committed buffer onto a checkpoint | |
616 | * list so that we know when it is safe to clean the transaction out of | |
617 | * the log. | |
618 | * | |
619 | * Called with the journal locked. | |
620 | * Called with j_list_lock held. | |
621 | */ | |
f7f4bccb | 622 | void __jbd2_journal_insert_checkpoint(struct journal_head *jh, |
470decc6 DK |
623 | transaction_t *transaction) |
624 | { | |
625 | JBUFFER_TRACE(jh, "entry"); | |
626 | J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh))); | |
627 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | |
628 | ||
de1b7941 JK |
629 | /* Get reference for checkpointing transaction */ |
630 | jbd2_journal_grab_journal_head(jh2bh(jh)); | |
470decc6 DK |
631 | jh->b_cp_transaction = transaction; |
632 | ||
633 | if (!transaction->t_checkpoint_list) { | |
634 | jh->b_cpnext = jh->b_cpprev = jh; | |
635 | } else { | |
636 | jh->b_cpnext = transaction->t_checkpoint_list; | |
637 | jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev; | |
638 | jh->b_cpprev->b_cpnext = jh; | |
639 | jh->b_cpnext->b_cpprev = jh; | |
640 | } | |
641 | transaction->t_checkpoint_list = jh; | |
642 | } | |
643 | ||
644 | /* | |
645 | * We've finished with this transaction structure: adios... | |
646 | * | |
647 | * The transaction must have no links except for the checkpoint by this | |
648 | * point. | |
649 | * | |
650 | * Called with the journal locked. | |
651 | * Called with j_list_lock held. | |
652 | */ | |
653 | ||
f7f4bccb | 654 | void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transaction) |
470decc6 DK |
655 | { |
656 | assert_spin_locked(&journal->j_list_lock); | |
657 | if (transaction->t_cpnext) { | |
658 | transaction->t_cpnext->t_cpprev = transaction->t_cpprev; | |
659 | transaction->t_cpprev->t_cpnext = transaction->t_cpnext; | |
660 | if (journal->j_checkpoint_transactions == transaction) | |
661 | journal->j_checkpoint_transactions = | |
662 | transaction->t_cpnext; | |
663 | if (journal->j_checkpoint_transactions == transaction) | |
664 | journal->j_checkpoint_transactions = NULL; | |
665 | } | |
666 | ||
667 | J_ASSERT(transaction->t_state == T_FINISHED); | |
668 | J_ASSERT(transaction->t_buffers == NULL); | |
470decc6 | 669 | J_ASSERT(transaction->t_forget == NULL); |
470decc6 | 670 | J_ASSERT(transaction->t_shadow_list == NULL); |
470decc6 DK |
671 | J_ASSERT(transaction->t_checkpoint_list == NULL); |
672 | J_ASSERT(transaction->t_checkpoint_io_list == NULL); | |
a51dca9c | 673 | J_ASSERT(atomic_read(&transaction->t_updates) == 0); |
470decc6 DK |
674 | J_ASSERT(journal->j_committing_transaction != transaction); |
675 | J_ASSERT(journal->j_running_transaction != transaction); | |
676 | ||
2201c590 SA |
677 | trace_jbd2_drop_transaction(journal, transaction); |
678 | ||
470decc6 | 679 | jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid); |
470decc6 | 680 | } |